We put engineers on your site — underground, on the deck, on the slab — to watch how the work actually gets done. Then we tell you where AI belongs, and where it has no business being.
Software went where the data already was. Everything else got a pilot, a dashboard nobody opened, and a vendor who had never been on site. The bottleneck was never the models — it is that almost no one has stood next to a crew long enough to know which ten minutes of their shift is worth automating, and which ten minutes must never be touched.
The kind of work we go looking for:
Handover notes written on paper, retyped into two systems, and still incomplete. Maintenance history that lives in a supervisor's head until he retires. Assay reconciliation done by eye because the tooling never fit the assay.
A plan states what to build, never in what order. RFIs take four days because the answer sits in one inbox. Regulatory compliance is checked by a human re-reading the regs, on every job, forever.
Catch logged on a wet clipboard and keyed in days later. Quota reconciled in a spreadsheet nobody trusts. Cold-chain gaps discovered by the buyer rather than on the deck, when it is far too late to act.
Forward Deployed Engineers, in PPE, through your induction, following a crew for a full shift. Not a discovery workshop. Not a questionnaire. The things worth automating are almost never the things people report in a meeting.
Boots, PPE, site induction. We arrive on your terms and under your safety officer.
Day 1We follow the actual work, shift start to shift end, and write down what really happens.
Days 1–2Every task ranked: worth automating, not worth it, or dangerous to automate. In writing.
Week 1If something is worth building, we build the smallest version that proves it — in your environment.
Weeks 2–8Your team owns what we built. We keep what we learned about the shape of the problem.
OngoingHammerscrape is not a consultancy that bills by the hour until the money runs out. Every site we visit is a search for a problem that repeats — the same broken thing in every mine, on every site, on every boat. When we find one that repeats often enough, we build a product for it and stop selling the hours.
An open format for buildable plans. A drawing says what to build but never in what order, under which regulation, or where a human must be asked. Datum encodes all three so an agent can reason about a building instead of merely rendering it.
You cannot automate a building site while the plans are pictures. The format has to exist before the agents that consume it are worth writing. So we are proposing it publicly, before we have any commercial reason to own it.
We have no proposal for these yet, and will not invent one from a desk. The next candidates come out of the next sites we stand on.
We have no customers. No case studies, no logos, no deployment numbers — and we would rather tell you that than dress a pilot up as a track record. What we have is a method we believe in, and one concrete technical proposal published in full for you to pick apart. If that isn't enough to earn two days of your time, then it shouldn't be.
You give us PPE, an induction, and permission to follow a crew. We give you a written account of what we saw, and where we think AI would and would not help. No deck, no invoice, no obligation to do anything with it.